Wide band communications over cable is bread and butter to the xDSL networking community. xDSL achieves the bandwidth over POTS (plane old telephone system) by using sophisticated encoding schemes with error correcting capacity, dynamic equalization, and echo cancellation. An xDSL connection begins with a modem to modem conversation/initialization culminating in a continuous, long lasting point-to-point, full duplex sessions with modulated signals traveling in both directions all the time..
The bandwidth needs projected for the DOM twisted pairs come within a factor of two or three of the best xDSL data rates. One might assume that xDSL might, therefore, be a good technology for IceCube. The model for xDSL doesn't match up with the model for IceCube. Communications between DOMs and counting house data acquisition front end is regularly and often interrupted by timing synchronization signals. Communications must start and stop quickly to accommodate the alternate interrogation of a pair of DOMs sharing a twisted pair with a counting house electronics channel. The line must go silent to insure the lowest noise environment for the highest accuracy of time synchronizing. Furthermore, xDSL chip sets currently consume nearly the same amount of power as is the goal for the power budget for the entire DOM. A factor of 2 improvement in power consumption can be anticipated, but would still probably not be good enough.
During the January 2001 south pole visit the communications waveforms from DOMs in the ice to testboards were gathered.
The results are measured after a gain stage in the line receiver
circuitry, so they include the noise contributed by the
amplifier.
 |
The 600 kilobaud signal was an NRZ type encoding scheme with
a run length limited to 5 baud periods. The modulation is also
designed to be DC balanced over any pair of characters. [In fact,
it was a broken sample 8b10b from the Xilinx Inc. web site. The
run lengths were not limited to 5, especially across character
boundaries. Therefore, the DC balance condition was not well
satisfied in the short term, but was over many characters. This
15,000 points of this waveform were captured by the oscilloscope, and
emailed from the pole to Berkeley where some digital filtering was
applied. The faults in the encoder stood out even more than in
this screen shot.] The measurements on channel 1, 2, and 4 of the oscilloscope are the 'far-end' cross-talk of the signal driving the the input to oscilloscope channel 3. The far-end cross-talk appears to be at least 40 dB to 50 dB below the received signal... perhaps more. Regrettibly, averaging could not be employed during this test to suppress the noise so that a more accurate assessment of the cross-talk could be recorded. That will be left to future measurements, if they prove nessary... |
The simulation for the
exising compensation as well as other analog compensations (in the
signal path to the comparator chip in the DOM and test board design)
shows the effects of various component choices in an environment where
the rise time can vary due to cable lengths that can differ by as
much as 0.6 km for bore holes near and far from the counting house.
The RZ communications encoding
for String 18 can serve the needs of IceCube.